Experiment Velocity

Most organisations do not have an idea shortage. They have a per-test cost that nobody has ever measured, and that cost silently decides both how many experiments run and which ideas are allowed to be one. This skill is about the machine that produces results — the platform, the tooling, the self-serve surfaces and the standard readout — not about whether any single result is true.

Scan the descriptions, follow only the links you need.

The economics

• [[references/marginal-cost-decides-which-ideas-get-tested]] — volume tracks the cost of one more test, and expensive tests quietly restrict the portfolio to ideas someone already believes.
• [[references/experiments-as-shared-infrastructure]] — template setup, eligibility, hypothesis and nightly analysis into one internal service so learning rate scales with traffic rather than headcount.

Get the bottleneck out of the way

• [[references/take-the-domain-owner-off-the-engineering-queue]] — when the person with the judgement is not the person with commit access, a whole class of experiment never gets proposed.
• [[references/self-serve-adoption-is-the-tooling-metric]] — count the non-analysts who get their own answers, because every unanswerable question becomes a queued request or a silent guess.

What the machine should produce